Due to recent advances in sequencing technology, an increasing number of dense marker maps and fully sequenced genomes is becoming available for many populations in animal and plant breeding. In this thesis we study recently developed methods for the estimation of breeding values based on genomic data. In the first part of the thesis, we present models which estimate genetic effects of markers, hence genetic breeding values, based on phenotypic records from single traits. These models differ in their assumptions on the genetic architecture of the trait, i.e., on the number of QTLs and on their effect-sizes. Although the differences in accuracy between the models are smaller than expected, we determine a strong model-dependent influence of the genetic architecture on the accuracy of breeding values. We show that Bayesian models usually perform better than linear mixed models if a few QTLs determine the trait, whereas the opposite may be true if many QTLs are underlying the trait. Further, we explore the influence of the density of markers, the heritability of the trait, and the number of phenotypic records on the performance of the methods. In the second part of the thesis, we review multi-trait models. These models use the available data more efficiently than single-trait models by incorporating correlations between traits. The multi-trait models increase the accuracy of breeding values for low-heritability traits which are correlated to high heritability traits. Especially, if phenotypic records are missing for low-heritability traits, the use of multi-trait models is strongly recommended.